Decision making on optimization method

Hello Everyone!

I am currently working on a Simulink model for which I want to fit parameters using the Simulink Design Optimization Toolbox (2015b). Since I need to fit quite a few parameters for long lasting simulations I was wondering which optimization method to use in advance.

The design variables are restricted by bounds and do use the sum squared error as cost function. Thus its a Minimization problem.

I read the users guide (<http://www.apmath.spbu.ru/ru/staff/smirnovmn/files/sldo_ug.pdf>) but didnt find other sources.

I understand: - I cant use Simplex pattern due to the bounds on the design variables

I dont understand: - how to choose from the others since all of them appear to be usable to me.

So my question is how to decide which method I should use. Is there any document which shows the reasoning (benefits and drawbacks) in a handy way (for one that doesnt have mathematical background)?

Thanks in Advance!

Best Regards, Stephan

Answer 1

[Prateek Khandelwal]

I think the section "Optimization Methods and Problem Formulations" in the documentation of Simulink Design Optimization would help you.

[Stephan Heidrich]

Hi Prateek,

thank you for your answer, but unfortunately the website is quite redundant with the users guide i linked above. I was looking for some more detailed explanation.

Answer 2

[Alan Weiss]

For general optimization there are some general guidelines on which solver to choose. For a sum of squares, you should likely choose lsqnonlin, unless your objective function always outputs the sum of squares already calculated, in which case choose fmincon ( lsqnonlin wants to raw components, and internally sums the squares). Be careful, when optimizing a simulation you often have to choose larger-than-default finite differences.

If your objective function is not a smooth function of the control variables, and if you have a Global Optimization Toolbox license, then you might want to try patternsearch.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation